Stockfish Testing Queue

Finished - 40825 tests

17-08-18 pb0 knight_fork diff
LLR: -3.35 (-2.94,2.94) [0.00,5.00]
Total: 45004 W: 8328 L: 8276 D: 28400
sprt @ 10+0.1 th 1 Bonus for knight threatening fork on King and major piece
17-08-18 And statBonusByQuietCount2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 10857 W: 1697 L: 1777 D: 7383
sprt @ 10+0.1 th 1 Take 2. No cap on bonus. Add a malus component.
17-08-18 sg passed_pawns diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 47333 W: 8778 L: 8699 D: 29856
sprt @ 10+0.1 th 1 Only midgame bonus.
17-08-18 IIv tune_tmm diff
1501/60000 iterations
3131/120000 games played
120000 @ 10+0.1 th 1 Tune non-moves-to-go case of time management.
17-08-18 And statBonusByQuietCount diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 43433 W: 7138 L: 7090 D: 29205
sprt @ 10+0.1 th 1 Stats bonus for best quiet move, based on how many bad quiets were tried first. Capped at 2*depth.
17-08-18 Voy lmrEscapeCaptureT diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 11774 W: 2119 L: 2194 D: 7461
sprt @ 10+0.1 th 1 stc
17-08-18 pb0 pick_1_or_2 diff
ELO: 1.19 +-2.3 (95%) LOS: 84.4%
Total: 30000 W: 5239 L: 5136 D: 19625
30000 @ 5+0.05 th 5 Like to test more accurately if there's a measurable difference in quality between threads of different skipsizes. Take 1: test branch picks best thread among idx 1 & 2 (skipsize 1 pattern) whilst base branch among idx 3 & 4 (skipsize 2 pattern).
17-08-18 Voy lmrt diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 20325 W: 3723 L: 3761 D: 12841
sprt @ 10+0.1 th 1 stc
17-08-18 sg lmr_stats diff
LLR: -3.01 (-2.94,2.94) [0.00,5.00]
Total: 18406 W: 3445 L: 3493 D: 11468
sprt @ 10+0.1 th 1 Scaling with a maximum 20% difference up to depth 16
17-08-18 jos adaptive_pruning diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 8714 W: 1546 L: 1634 D: 5534
sprt @ 10+0.1 th 1 Take 2, based on material. (Bugix, thanks Stefan!)
17-08-18 pb0 limit_threats diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 7401 W: 1297 L: 1391 D: 4713
sprt @ 10+0.1 th 1 Limit threats scoring by minors/rook to max 2 times per type: as we can just capture only piece per move, multi-tip forks might produce an exaggerated score.
17-08-18 jos adaptive_pruning diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 1601 W: 252 L: 372 D: 977
sprt @ 10+0.1 th 1 Take 2, based on material.
17-08-18 Elb npm_r diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 4345 W: 747 L: 854 D: 2744
sprt @ 10+0.1 th 1 More nmp reduction in middlegame, less in endgame
17-08-18 Voy lmrt3c diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 47666 W: 8706 L: 8627 D: 30333
sprt @ 10+0.1 th 1 Try a bigger delta (20k)
17-08-18 Fis memShrink diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 25458 W: 4586 L: 4656 D: 16216
sprt @ 10+0.1 th 1 Only keep the most significant change according to bench which is the pawn hash.
17-08-18 fau QueenFauzi diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 7198 W: 1278 L: 1373 D: 4547
sprt @ 10+0.1 th 1 First try for a new idea, (QueenValue -100 , QueenAdjust +200) (first guessed numbers)
17-08-16 And researchStatBonus4A diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 75210 W: 12416 L: 12242 D: 50552
sprt @ 10+0.1 th 1 Add depth to the equation. (Park with prio -1 until researchStatBonus4 fails)
17-08-17 pro ps_sort_custom diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 39063 W: 7011 L: 6971 D: 25081
sprt @ 10+0.1 th 1 stage 2: converging to ~(128<<((time>>7)+depth)
17-08-17 And researchStatBonus4B diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 82295 W: 13532 L: 13331 D: 55432
sprt @ 10+0.1 th 1 STC : Same as researchStatBonus4, but double the bonus and limit it to depths at or below twelve, similar to stat_bonus(). May I stop researchStatBonus4A?
17-08-16 Voy lmrQS3 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 85786 W: 15426 L: 15189 D: 55171
sprt @ 10+0.1 th 1 Ver. 3
17-08-17 Fis memShrink diff
LLR: -2.94 (-2.94,2.94) [0.00,4.00]
Total: 66130 W: 11871 L: 11803 D: 42456
sprt @ 10+0.1 th 1 Now that stats16 is in let's test this for a gain.
17-08-17 Voy lmrQSv2 diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 50172 W: 9107 L: 9019 D: 32046
sprt @ 10+0.1 th 1 Take 2
17-08-17 Voy lmrQSv3 diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 21706 W: 3851 L: 3884 D: 13971
sprt @ 10+0.1 th 1 Fixed a silly bug...
17-08-17 Voy lmrQSv3 diff
LLR: -1.88 (-2.94,2.94) [0.00,5.00]
Total: 9613 W: 1697 L: 1736 D: 6180
sprt @ 10+0.1 th 1 stc
17-08-17 vdv scorePromo diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 23861 W: 4264 L: 4288 D: 15309
sprt @ 10+0.1 th 1 stc take 2
17-08-17 mco stats16 diff
ELO: -8.58 +-10.2 (95%) LOS: 5.0%
Total: 1175 W: 141 L: 170 D: 864
2000 @ 60+0.6 th 1 Detect overflow in history (engine will crash!). Take 2 at LTC to have more time for stats to fill up.
17-08-17 sg passed_pawns diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 28945 W: 5209 L: 5211 D: 18525
sprt @ 10+0.1 th 1 Only endgame bonus
17-08-17 sg lmr_stats diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 32410 W: 5877 L: 5864 D: 20669
sprt @ 10+0.1 th 1 Even less scaling (maximum different of 10% up to depth 16)
17-08-17 vdv scorePromo diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 18200 W: 3181 L: 3230 D: 11789
sprt @ 10+0.1 th 1 stc against #1201
17-08-17 jos adaptive_pruning diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 1491 W: 225 L: 345 D: 921
sprt @ 10+0.1 th 1 First draft of adaptive pruning based on number of legal moves.
17-08-17 pb0 smp_pickbest3 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 20486 W: 3389 L: 3430 D: 13667
sprt @ 5+0.05 th 5 Take 2: Ignore selective depth, just replace completedDepth with nominalDepth (=effective used rootDepth for best move).
17-08-17 sg passed_pawns diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 5773 W: 966 L: 1067 D: 3740
sprt @ 10+0.1 th 1 Double bonus
17-08-17 sg update_stats_mcp diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 44946 W: 8134 L: 8068 D: 28744
sprt @ 10+0.1 th 1 depth < 5
17-08-17 sg passed_pawns diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 40818 W: 7449 L: 7400 D: 25969
sprt @ 10+0.1 th 1 Small additional passed pawn bonus in pawns.cpp to so also in lazy eval passed pawns are recognized.
17-08-17 sg lmr_stats diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 5224 W: 887 L: 990 D: 3347
sprt @ 10+0.1 th 1 Less scaling (Park with prio -1)
17-08-17 sg lmr_stats diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 3836 W: 608 L: 716 D: 2512
sprt @ 10+0.1 th 1 For LMR scale stat score by depth
17-08-17 And researchStatBonus4 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 15961 W: 1763 L: 1833 D: 12365
sprt @ 60+0.6 th 1 LTC
17-08-17 pro ps_sort_custom diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 24656 W: 4430 L: 4450 D: 15776
sprt @ 10+0.1 th 1 stage 2: dynamically calculate sort limit: 1024 - 15*depth*optimumTime
17-08-15 And researchStatBonus4 diff
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 151824 W: 25179 L: 24459 D: 102186
sprt @ 10+0.1 th 1 STC Take 4. Scale with R value from LMR instead of node depth.
17-08-16 Voy lmrQSv diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 11190 W: 1408 L: 1490 D: 8292
sprt @ 60+0.6 th 1 LTC: Different variant...
17-08-16 Voy lmrQSv diff
LLR: 3.56 (-2.94,2.94) [0.00,5.00]
Total: 40696 W: 7416 L: 7094 D: 26186
sprt @ 10+0.1 th 1 Different variant... (fixed patch)
17-08-16 sg lazy_eval2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 19574 W: 3421 L: 3464 D: 12689
sprt @ 10+0.1 th 1 Same thresholds (1500)
17-08-16 pro ps_sort_zeros diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 16192 W: 2895 L: 2951 D: 10346
sprt @ 10+0.1 th 1 1024<<depth when depth < 5, otherwise sort all at 8-bit resolution.
17-08-16 sg lazy_eval2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 14001 W: 2474 L: 2540 D: 8987
sprt @ 10+0.1 th 1 Higher threshold (1350)
17-08-16 IIv pieceValues diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 8798 W: 1526 L: 1653 D: 5619
sprt @ 10+0.1 th 1 Take 3.
17-08-16 pro ps_sort_zeros diff
LLR: -1.38 (-2.94,2.94) [0.00,5.00]
Total: 33280 W: 5849 L: 5769 D: 21662
sprt @ 10+0.1 th 1 limit = 1024<<depth, for depth < 5. otherwise, sort everything roughly.
17-08-16 sg lazy_eval2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 3823 W: 639 L: 748 D: 2436
sprt @ 10+0.1 th 1 Add second early lazy eval exit after mobility
17-08-16 pb0 smp_pickbest3 diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 19688 W: 2704 L: 2756 D: 14228
sprt @ 20+0.2 th 3 LTC: Trying to further improve pickbest-logic by comparing nominalDepth (=effective used rootDepth for best move) instead of completedDepth.
17-08-16 sg lazy_eval_scale2 diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 9357 W: 1649 L: 1734 D: 5974
sprt @ 10+0.1 th 1 Scale lazy eval only in endgame.
17-08-16 sg lazy_eval_scale diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 4127 W: 666 L: 773 D: 2688
sprt @ 10+0.1 th 1 If lazy eval is triggered return instead the value after standard game phase formula (with scaling factor). The idea is that some drawisch endgames only detected with the scale factor which in the current formula would be ignored and possible overoptimistic evals are returned.